Freezing the acceptance probability past the halting time #
A probabilistic machine whose paths all halt within T steps has the same
acceptance probability at every later time bound: the extra choice bits are
read by no transition, so they only refine the sample space uniformly. This
lets a machine's arbitrary time-bound function be replaced by a polynomial
that dominates it, which is what makes the Lautemann matrix predicate
computable — see
Complexitylib.Classes.PH.SipserLautemann.Matrix.
Main results #
card_filter_blockFst_eq— a prefix fiber of the seed space has2 ^ bpointsNTM.acceptProb_eq_of_allPathsHaltIn— the acceptance probability is frozen past the halting timeNTM.acceptsWithProb_of_le,NTM.rejectsWithProb_of_le— the bounded-error conditions transfer to any pointwise-larger time bound
The acceptance probability is frozen past the halting time. If all
paths halt within T (|x|) steps, running the machine for any longer bound
leaves the acceptance probability unchanged: the surplus choice bits partition
the enlarged sample space into equal fibers over the original one.
The completeness condition transfers to any pointwise-larger time bound.
The soundness condition transfers to any pointwise-larger time bound.